Tag
A research paper finds that base language models appear human to AI detectors, unlike instruction-tuned models. The authors propose a paraphrasing pipeline (HIP) that improves human-likeness while preserving semantics across model sizes.